Thin vs. Unicorn
Update: There is a discussion page over at HN now.
I’ve been hearing a lot of buzz lately about a new Ruby web application server called Unicorn. It can run Rack applications (including Ruby on Rails, of course), and it has been engineered to tie in very closely with the operating system to boost performance.
Thin is another web application server that’s somewhat similar to Unicorn in that it can also run Rack applications. Thin uses the EventMachine library, however, which gives it a big speed boost over the previous standard application server, Mongrel. EventMachine also enables it to be used for asyncronous things like long-polling Comet servers.
Benchmarking
I needed to choose a web server for Pushy, and I initially chose Thin. I thought it would be interesting, however, to see which of Thin and Unicorn was faster. My benchmarks were very simple, and I used the following rackup file to test the servers:
class HelloRack
def call(env)
[
200,
{ "Content-Type" => "text/html" },
["Hello, Rack!"]
]
end
end
run HelloRack.new
Pretty simple, right? In addition, I used the following command to benchmark:
ab -n 10000 -c 1000 http://127.0.0.1:3000/
The Results
Thin, one process: 3638 requests / second
Thin, three processes: 4347 requests / second
Unicorn, one worker: 6774 requests / second
Unicorn, three workers: 8,012 requests / second
Conclusion
Clearly, Unicorn gives some nice performance benefits. It also seems a lot more flexible, as you don’t even need to use something like nginx to load balance between worker processes — it ties in directly with the operating system and load balances itself, which obviously lent to the performance of it.
As a disclaimer, I realise that this isn’t a complex benchmark at all, and there are probably ways that it could be made to be more accurate. I believe that it will give people that had the same question that I did an idea of which web application server might perform better for them.
